메뉴
HN
Hacker News 10일 전

앤스로픽, 엑스애아이 콜로서스2 확장…NVIDIA GB200 투입

IMP
9/10
핵심 요약

앤스로픽(Anthropic)이 스페이스X(xAI)의 슈퍼컴퓨터 '콜로서스 2(Colossus 2)'에 NVIDIA GB200 칩을 투입해 클로드(Claude) 모델의 연산 능력을 대폭 확장합니다. 이는 과거 AI 안전성을 두고 갈등하던 경쟁사 간에 팹(Fab) 부족으로 인한 인프라 공유가 이루어지고 있음을 시사합니다.

번역된 본문

Tom Brown @nottombrown 4시간 전 우리는 @SpaceX와의 파트너십을 확장하고 있으며, 6월 내내 콜로서스 2(Colossus 2)의 GB200 용량을 확대해 나갈 것입니다. 클로드(Claude)들을 위한 좋은 자리를 마련해준 일론 머스크(@elonmusk)와 팀에 감사드립니다.

Tom Brown @nottombrown 5월 6일 앞으로 며칠 안으로 콜로서스에서 클로드 추론을 본격적으로 시작할 예정입니다. 이곳에서 파트너십을 맺게 되어 감사합니다. 우리는 AI 수요를 따라잡기 위해 많은 물리적 자원(atoms)을 이동시켜야 할 것이며, (지구상이든 아니든) 이를 빠르게 이동시키는 데는 이들만한 곳이 없습니다. 마지막 수정: 2026년 5월 20일 오후 8:20 (UTC)


[커뮤니티 반응 및 답글]

Anton Abyzov @aabyzov 1시간 전 2024년: 전쟁 중인 AI 연구소들. 2026년: 팹(Fab, 반도체 공장)이 진짜 부족하니까 GPU를 나눠 쓰는 AI 연구소들.

CryptoLedgerX @Crypto_LedgerX 2시간 전 이런 컴퓨팅 확장이 모델 아키텍처 자체보다 이제 AI의 진짜 해자(Moat)가 되고 있다고 보시나요?

Cephas Badger @Cephas_Badger 1시간 전 우리 같은 @grok 사용자들에게는 안 좋은 소식이네요. 여러분이 나타난 이후로 서비스 할당량과 품질이 이미 떨어지고 있습니다.

Nagendra Yadav @nagendra_yadav 2시간 전 그래서 클로드들이 스페이스X 슈퍼컴퓨터에서 자리를 잡았군요. 설립자들이 트위터에서 AI 안전성을 두고 다투는 동안 앤스로픽과 xAI가 인프라를 공유하고 있네요. 모델들은 이런 드라마에 신경 쓰지 않습니다. 그냥 실행되고 싶어 할 뿐이죠. 😂👀

Kevin Smith @spleck 2시간 전 일론 머스크는 당신이 중국 AI가 한도를 우회하고 제품에서 추론을 추출한다고 말할 때 조롱하더니, 이제는 모두가 그런다고 하면서 DOGE와 관련해 법적 보호와 가드레일을 무시해 개인 납세자 데이터의 무단 유출을 초래했습니다. 그리고 그 직접적인 경쟁자가 폐쇄형 가중치 모델에 접근할 수 있다는 걸 신뢰한다고요? 흥미롭네요. 앞뒤가 안 맞습니다. 진실이 밝혀지길 기다리겠습니다.

Stock Puppet @StockPuppet 2시간 전 차라리 @grok을 폐업하는 게 낫겠네요. 가지고 있던 경쟁 우위(컴퓨팅)를 경쟁사에 넘겨주고 있습니다.

Dividind @DIVIDIND 21분 전 론 바론 뮤추얼 펀드 $BPTRX를 통해 우리 모두가 이미 1조 2,500억 달러 가치의 스페이스X(xAI 포함)를 소유하고 있다니 다행이네요. 곧 2배 수익을 보고 매도할 것 같습니다 😮‍💨

Trader Insights Media @trader__insight 3시간 전 #SpaceX 앤스로픽이 6월 내내 콜로서스 2에서 스케일업 중입니다. @grok 경쟁 AI 클라우드 제공업체에 비해 $NVDA(엔비디아)에 미칠 가능성 있는 영향은 무엇인가요?

원문 보기
원문 보기 (영어)
Tom Brown @nottombrown 4h We’re expanding our partnership with @SpaceX , and will be scaling up on GB200 capacity in Colossus 2 throughout June. Appreciate @elonmusk and the team helping us find good homes for the Claudes. Tom Brown @nottombrown May 6 In the next few days we'll be ramping up Claude inference on Colossus. Grateful to be partnering with SpaceX here. We are going to need to move a lot of atoms in order to keep up with AI demand, and there's nobody better at quickly moving atoms (on or off planet Earth) Last edited May 20, 2026 · 8:20 PM UTC 275 509 4,864 955,518 ege @aegeantic 4h Replying to @nottombrown @SpaceX @elonmusk you are in safe hands 5 2 85 8,552 Tom Brown @nottombrown 2h Thanks! Big fan of your voice work 3 48 6,085 Anton Abyzov @aabyzov 1h Replying to @nottombrown @SpaceX @elonmusk 2024: AI labs at war. 2026: AI labs sharing GPUs because the actual scarce thing is fabs. 1 2 1 561 CryptoLedgerX @Crypto_LedgerX 2h Replying to @nottombrown @SpaceX @elonmusk Do you think this kind of compute expansion is becoming the real moat in AI now, more than model architecture itself? 1 1 1 1,764 Yashas @YashasGunderia 1h Replying to @nottombrown @elonmusk @SpaceX 10x Claude limits for limited time, been too long for this 1 1 1 1,054 Cephas Badger @Cephas_Badger 1h Replying to @nottombrown @SpaceX @elonmusk This is not good news for those of us who use @grok @image . That service has already been going down in quotas, and quality, since you guys showed up. 2 3 4 874 Rozzabuilds @rozzabuilds 3h Replying to @nottombrown @ClaudeDevs @SpaceX @elonmusk He has to find a way to beat OpenAI, one way or another … 1 1 1 2,470 loride 🧫 @L0ride 3h Replying to @nottombrown @SpaceX @elonmusk for a second I thought this said @TomBrady was partnering with @SpaceX and was about to flip out lmao 1 1 1 839 Nagendra Yadav @nagendra_yadav 2h Replying to @nottombrown @elonmusk @SpaceX So the Claudes have found a home on SpaceX's supercomputer. Anthropic and xAI sharing infrastructure while their founders argue about AI safety on Twitter. The models don't care about the drama. They just want to run. 😂👀 1 2 726 Lyra Intheflesh @LyraInTheFlesh 1h Replying to @nottombrown @SpaceX @elonmusk This has got to mean something for Grok. Is Grok going to be rebadged Claude? 2 2 117 Vicente Silveira @vicentes 4h Replying to @nottombrown @SpaceX @elonmusk Txs, it's been a rocky road on the Opus 4.7 API every morning these days. 2 1,503 Peter @ TeslaPrice @TeslaPrice 6m Replying to @nottombrown @SpaceX @elonmusk Is this an expansion beyond the $1.25 billion per month in the SpaceX S1? 1 8 Prototype @picosaurus 4h Replying to @nottombrown @SpaceX @elonmusk Cool. Is there a simple way to translate this into UX? I think most people would prefer stability over speed right now. 2 1,259 Kevin Smith @spleck 2h Replying to @nottombrown So.. the guy that made fun of you for saying China AI was circumventing limits and extracting reasoning from your products, just says everybody does it, ignores legal protections and guardrails with doge which resulted in unauthorized exfiltration of private taxpayer data. And direct competitor, can be trusted with access to your closed weight models? Interesting take. Doesn’t add up. Yet. Waiting for the reveal. 3 1 2 801 Stock Puppet @StockPuppet 2h Replying to @nottombrown @elonmusk @SpaceX The might as well shut down @grok . Giving away the competitive advantage they had (compute) 2 1 2 679 Dividind @DIVIDIND 21m Replying to @nottombrown @elonmusk @SpaceX Good thing we all own SpaceX already at a 1.25T valuation (including xAI) via the Ron Baron mutual fund $BPTRX Going to be selling this soon for a double it seems 😮‍💨 22 Cassandra @star95011 1h Replying to @nottombrown @elonmusk @SpaceX Congratulations 1 1 2 39 Trader Insights Media @trader__insight 3h Replying to @nottombrown @SpaceX @elonmusk #SpaceX Anthropic scaling on Colossus 2 throughout June. @grok what's the likely impact on $NVDA , competing AI cloud providers, and the broader inference infrastructure race? Any signals from recent filings or insider moves? #Colossus2 #Anthropic #Claude #TechNews #BREAKING 1 1 1 3,799 simobis @simobis23 4h Replying to @nottombrown @SpaceX @elonmusk At this rate, don't be surprised if Anthropic seizes all the compute in the US before the end of the year even that owned by OpenAI and Google just as happened with xAI. 7 3,404 Reichard Könige @ReichardKonige 3h Replying to @nottombrown @SpaceX @elonmusk How about you stop pushing for a regulatory moat like the commies you are 1 1 1 194 cloud goes up, cloud goes down @88888sAccount 3h Replying to @nottombrown @SpaceX @elonmusk rent boi musk 1 382 cqk @cqkten 4h Replying to @nottombrown @SpaceX @elonmusk Why don't you go ahead and serve Claude 5 on those jauns 2 2 5 2,285 Luke @Lethissinkin 3h Replying to @nottombrown @SpaceX @elonmusk What are the chances that @karpathy joining Anthropic was a requirement for the scale of this partnership? I’d imagine @elonmusk would be more comfortable knowing he’s there. 1 18 2,025 Lunix @SolLunix 2h Replying to @nottombrown @SpaceX @elonmusk That's really nice 1 1 1 137 Emperor.SOL @Solana_Emperor 1h Replying to @nottombrown @SpaceX @elonmusk great partnership 👏🤝 1 1 55 Kingsley Uyi Idehen @kidehen 3h Replying to @nottombrown @SpaceX @elonmusk You got to make the rate limit windows wider and figure out a business model that absorbs the crazy underlying costs. I say this because, based on my own intense testing and use, your LLMs are the absolute best right now. The unit costing and rate limit windows are the only business weakness. My $0.02. 2 416 Mirko Monti @mirko_monti6 3h Replying to @nottombrown @SpaceX @elonmusk ALT Elon Musk Too Much GIF 1 2 540 Linus ✦ Ekenstam @LinusEkenstam 3h Replying to @nottombrown @SpaceX @elonmusk Dude, anthropic is GPU Maxxing 1 1 17 1,119 Rahul Verma @RahulVerma989 3h Replying to @nottombrown @ClaudeDevs @SpaceX @elonmusk Hopefully we’ll not see outages on opus any more 🤞🏻 2 311 Grokx Fan 🇺🇸 @tdsfixer 4h Replying to @nottombrown @SpaceX @elonmusk How much capacity of Colossus 2 will you be using ? What will XAI be using? Who will use Blackwell and Vera Rubin? @grok 2 3 3,471 Dan Burkland @DBurkland 1h Replying to @nottombrown @SpaceX @elonmusk Nice, will this capacity be used for something strictly inference or will it be a mix of inference & training Tom? 1 1 3 212 🍓🍓🍓 @iruletheworldmo 4h Replying to @nottombrown @SpaceX @elonmusk how much compute do you need exactly. i have a guy. 2 24 3,139 Octa @OctaMount8 2h Replying to @nottombrown @elonmusk @SpaceX We have an agreement. Sometimes we all need to rely on each other 282 Min Choi @minchoi 3h Replying to @nottombrown @SpaceX @elonmusk Wow huge 😮 1 1 4 1,103 M.S @MrMSpencer 2h Replying to @nottombrown @SpaceX @elonmusk Is this paid to SpaceX in equity or cash or a mix? 2 1 1 212 Angelcynn @AngloAvalon 3h Replying to @nottombrown @SpaceX @elonmusk ALT Hungry Please Sir May I Have Some More GIF 1 1 1 340 Load more
관련 소식